一种邻域自适应半监督局部Fisher判别分析算法

For the discriminant analysis of multimodal data, the idea of localization can hardly reflect the difference of local geometric structure according to the global setting of local neighborhood by experience. Aiming at this problem, this paper proposed a neighborhood adaptive semi-supervised local Fisher discriminant analysis (NA-SELF) algorithm. The new algorithm based on the semi-supervised local Fisher discriminant analysis algorithm, obtained the initial neighborhood by combining the Mahalanobis distance and cosine similarity, and adjusted the number of neighbors according to the probability density estimation of sample space. The performance of feature dimensionality reduction using the algorithm was verified by the synthetic datasets and five UCI standard datasets. Compared with several typical dimensionality reduction algorithms and the discriminant analysis algorithm using the traditional k-nearest neighbor method, the experimental results show that the proposed algorithm has higher effectiveness.